Cost-Constrained feature selection in binary classification: adaptations for greedy forward selection and genetic algorithms

摘要

Feature selection is an important and widely applied preprocessing step in the field of biomarker detection. In high-dimensional data sets, which are often found in the “-omics” field (genomics, transcriptomics, proteomics, metabolomics), many input variables may not carry relevant information for a given task. Others may represent redundant information. Excluding these features from the model building process can drastically improve predictive power in such situations.